From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754525AbbHFPQD (ORCPT ); Thu, 6 Aug 2015 11:16:03 -0400 Received: from mga09.intel.com ([134.134.136.24]:1528 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753554AbbHFPQB (ORCPT ); Thu, 6 Aug 2015 11:16:01 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.15,622,1432623600"; d="scan'208";a="743498314" Date: Thu, 6 Aug 2015 23:25:15 +0800 From: Feng Tang To: Michal Nazarewicz Cc: Greg Kroah-Hartman , John Stultz , Andrew Morton , Kyungmin Park , Marek Szyprowski , Joonsoo Kim , linux-kernel@vger.kernel.org Subject: Re: [PATCH] staging: ion: Add a default struct device for cma heap Message-ID: <20150806152515.GA30274@shbuild888> References: <1438856698-21381-1-git-send-email-feng.tang@intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Michal, Thanks for the review! On Thu, Aug 06, 2015 at 01:40:28PM +0200, Michal Nazarewicz wrote: > On Thu, Aug 06 2015, Feng Tang wrote: > > When trying to use several cma heaps on our platforms, > > we met a memory issue due to that the several cma_heaps > > are sharing the same "struct device *". > > > > As in current code base, the normal cma heap creating > > process is, one platform device is created during boot, > > and it will sequentially create cma heaps (usually passing > > its own struct device * as a parameter) > > > > For the multiple cma heaps case, there will be one "struct > > cma" created for each cma heap, and this "struct cma *" is > > saved in dev->cma_area. So the single platform device can't > > meet the requirement here. > > > > So this patch add one default device for a cma heap to avoid > > sharing the same "struct device", thus fix the issue. And it > > doesn't break existing code by only using that default device > > when no "struct device *" is passed in. > > > > Also, since the cma framework has been cleaned up, this patch > > also add a platform data member to pass the "struct cma*" to > > ion_cma_heap_create(). > > > > Signed-off-by: Feng Tang > > >From CMA’s point of view: > > Acked-by: Michal Nazarewicz > > > --- > > drivers/staging/android/ion/ion.h | 4 ++++ > > drivers/staging/android/ion/ion_cma_heap.c | 20 +++++++++++++++++--- > > 2 files changed, 21 insertions(+), 3 deletions(-) > > > > diff --git a/drivers/staging/android/ion/ion.h b/drivers/staging/android/ion/ion.h > > index 443db84..e9af17e 100644 > > --- a/drivers/staging/android/ion/ion.h > > +++ b/drivers/staging/android/ion/ion.h > > @@ -45,6 +45,9 @@ struct ion_buffer; > > * @size: size of the heap in bytes if applicable > > * @align: required alignment in physical memory if applicable > > * @priv: private info passed from the board file > > + * @priv2: when creating CMA heap, platform device should better also > > + * pass the "struct cma *" info, so that the cma buffer request > > + * know where to go for the buffer > > * > > * Provided by the board file. > > */ > > @@ -56,6 +59,7 @@ struct ion_platform_heap { > > size_t size; > > ion_phys_addr_t align; > > void *priv; > > + void *priv2; > > Why are those void pointers anyway? Perhaps just make them struct device > *dev and struct cma *cma? Especially since priv2 is a bit awkward name. My initial thought is the same, but as there are several other kinds of ion heaps which are also using this structure for their own ion_xxx_heap_create(struct ion_platform_heap *), I gave up using the explicit "struct cma *", in case other kinds of heaps may need to use this additional priv2 in future > > + * data->priv for cma heap is currently supposed to point > > + * to a "struct device *" > > + */ > > + if (data->priv) { > > + cma_heap->dev = data->priv; > > + } else { > > + cma_heap->dev = &cma_heap->default_dma_dev; > > + cma_heap->dev->coherent_dma_mask = DMA_BIT_MASK(32); > > + cma_heap->dev->dma_mask = &dev->coherent_dma_mask; > > + } > > + > > + /* data->priv2 contains a pointer to struct cma */ > > + dev_set_cma_area(cma_heap->dev, data->priv2); > > Perhaps: > > + if (data->priv2) > + dev_set_cma_area(cma_heap->dev, data->priv2); Yes, this looks more logical, even though the cma_heap structure is kzalloced. Thanks, Feng